You are on page 1of 16

Research Policy 34 (2005) 305320

Universityincubator firm knowledge flows: assessing


their impact on incubator firm performance
Frank T. Rothaermela , Marie Thursbyb
a

College of Management, Georgia Institute of Technology, Atlanta, GA 30308-0520, USA


b Georgia Institute of Technology and NBER, USA

Received 20 June 2004; received in revised form 25 October 2004; accepted 29 November 2004
Available online 17 March 2005

Abstract
Technology incubators are university-based technology initiatives that should facilitate knowledge flows from the university
to the incubator firms. We thus investigate the research question of how knowledge actually flows from universities to incubator
firms. Moreover, we assess the effect of these knowledge flows on incubator firm-level differential performance. Based on the
resource-based view of the firm and the absorptive capacity construct, we advance the overarching hypothesis that knowledge
flows should enhance incubator firm performance. Drawing on longitudinal and fine-grained firm-level data of 79 technology
ventures incubated between 1998 and 2003 at the Advanced Technology Development Center, a technology incubator sponsored
by the Georgia Institute of Technology, we find some support for knowledge flows from universities to incubator firms. Our
evidence suggests that incubator firms absorptive capacity is an important factor when transforming university knowledge into
firm-level competitive advantage.
2005 Elsevier B.V. All rights reserved.
Keywords: Knowledge flows; Localized spillovers; Resource-based view of the firm; Absorptive capacity; Incubator firm performance

1. Introduction
How do technology ventures access university
knowledge and how does it affect their performance?
Knowledge produced in universities has been studied
extensively, as has its impact on industry. Yet, we know
Corresponding author. Tel.: +1 404 385 5108;
fax: +1 404 894 6030.
E-mail address: frank.rothaermel@mgt.gatech.edu
(F.T. Rothaermela ).

0048-7333/$ see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.respol.2004.11.006

little about knowledge flows at the firm level, either


in terms of the flows themselves or effects thereof. In
part, this is because of the inherent difficulty tracking knowledge created for the public domain, but in
part, it is because the firm has not been a common
unit of analysis. Moreover, there is mounting empirical evidence that local knowledge spillovers produced
by university research are not free, but depend on
contractual agreements. Thursby and Thursby (2002)
and Zucker and Darby (1996, 1998) document this.
Moreover, Cockburn and Henderson (1998) demon-

306

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

strate that firms must exhibit substantial absorptive capacity to capture and appropriate rents to publicly available knowledge. Cohen and Levinthal (1989) advance
the notion of absorptive capacity, which is understood
as a firms ability to recognize, value, and assimilate
new external information.
Herein, we attempt to address the two-pronged research question of (1) how knowledge flows from universities to incubator firms and (2) how these flows
affect the performance of new technology ventures. As
part of the first question, we identify and analyze the
effects of different mechanisms through which knowledge flows from universities to incubator firms: university license, and patent backward citations to university
research, academic journals, research by the incubatorsponsoring university, and research from other universities than the sponsoring university. Embedded in the
second research question is the search for an appropriate performance metric for nascent technology ventures, a significant methodological challenge, which
has clearly retarded empirical research in this important area as emphasized by Phan et al. (2004).
Given the dearth on empirical research investigating university knowledge flows and their effect on
incubator firm performance, we develop two explorative hypotheses that we subsequently examine econometrically. Firstly, we argue that exclusive knowledge
flows in terms of a university license can endow the
start-up with a unique resource. Important theoretical
work in the strategic management literature has argued
that valuable, rare, inimitable, and non-substitutable resources may endow a firm with a competitive advantage
that translate to superior performance (Barney, 1991).
Secondly, we suggest that university backward patent
citations are indicative of a start-ups absorptive capacity that enables it to recognize public knowledge flows
emanating from a university, assimilate them internally,
and then to apply them to commercial ends (Cohen and
Levinthal, 1989). This in turn should lead to a variance
in performance among technology ventures with a ventures absorptive capacity being positively correlated
with venture performance.
We test these two tentative hypotheses on a sample of 79 incubator firms incubated in the Advanced
Technology Development Center (ATDC) at the Georgia Institute of Technology (Georgia Tech (GT)). We
follow these firms over the 6-year time span between
1998 and 2003. The use of an annual repeat survey en-

ables us not only to collect fine grained data for nascent


technology ventures, but also aids us in overcoming a
survivor bias common to research on new technology
ventures. Moreover, we attempt to enhance the robustness of the analysis by applying four different performance measures: total revenues, total funds obtained,
venture capital (VC) funding obtained, and failure or
graduation from the incubator. Applying different outcome variables might aid in identifying an appropriate
outcome measure in the new venture context, in particular, when attempting to capture the performance
implications of university knowledge flows.
This paper has the following outline. Section 2 reviews prior research on universityindustry knowledge
flows, and Section 3 develops the hypotheses regarding
university licenses, patent citations, and new venture
performance. Section 4 discusses the methodology applied, Section 5 presents the empirical results, while
Section 6 concludes this paper with a discussion of the
results, limitations as well as implications for future
research and public policy.

2. Universityindustry knowledge-owsprior
research
Early work on the industrial impact of academic research includes Adams (1990), who showed that academic knowledge, as measured by publications, was
a major contributor to productivity growth for 18 of
20 two-digit U.S. manufacturing industries from 1943
to 1983, albeit with a substantial lag times which varied from 010 years for applied sciences and engineering to 20 years for basic science publications. Jaffe
(1989) classic study of the real effects of academic research showed that university research had significant
effects on the generation of industrial patents at the state
level.
With the exception of the work by Zucker and Darby
(1996, 1998) and Zucker et al. (2002), which we discuss
below, the focus of research in this area is not overall
firm performance but the effect of university research
on industry R&D output. Even when data were collected by firm, the questions of interest have been variations in the relevance of university research by industry
and academic field. For example, both the 1983 Yale
Survey and the 1994 Carnegie Mellon Survey of R&D
managers asked the relevance of university research

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

for technical progress in their industry (Klevorick et


al., 1994; Cohen et al., 1998). Mansfields survey of
R&D executives of 66 firms examined the perceived
impact of university research on the firms ability to
develop new processes and products in a timely fashion (Mansfield, 1995). Not surprisingly, all of this work
finds the most pervasive effects of university research
are in the drug, chemical, and electronics industries.
This prior research demonstrates that knowledge
flows from universities tend to be mitigated by geographic distance, which of course suggests that academic publications alone cannot be the sole means by
which firms gain access to university knowledge. Researchers have used a variety of methods to examine the
channels by which university knowledge is transferred
to industry, including interviews and survey research,
citations to academic publications and patents, collaboration patterns in academic publications and patents,
as well as information on formal contracts such as licenses or sponsored research. The most comprehensive
survey in this regard is the Carnegie Mellon Survey of
1478 R&D laboratories which asked R&D managers
the importance to them of 10 channels of knowledge
flow (patent, publications, meetings or conferences, informal channels, hires, licenses, joint ventures, contract
research, consulting, and personal exchange). Of these
publications, public meetings and conferences, informal and personal information channels, and consulting
contracts appear to be the four most important channels, suggesting a complementarity between publication and other mechanisms involving personal contact
(Cohen et al., 1998). The results on consulting support
the results from Mansfield (1995) that show a complementarity between consulting and the research agendas
of university scientists working with industry.
The use of citation data, either to academic publications or patents, in economics has a rich tradition,
not only for examining knowledge flows from universities, but also R&D spillovers in general (Griliches,
1992). With regard to flows from universities, a series
of important studies examine a variety of measures of
citations to university patents in order to examine issues of localization as well as the importance of university patents (Jaffe et al., 1993; Trajtenberg et al., 1997;
Henderson et al., 1998). Branstetter (2004) examines
patent citations to academic publications and shows an
increasing trend for industrial patents to cite academic
science. This work complements that of Agrawal and

307

Henderson (2002) and Murray and Stern (2004) which


examines the interaction of academic and industry collaboration and citation patterns. Here, one underlying
assumption is that knowledge spillovers tend to be localized, and that being located close to the knowledge
source enhances the efficiency and effectiveness of the
knowledge transfer. When studying the performance
of U.S. university research parks, for example, Link
and Scott (2004) found that parks located closer geographically to a research university grew significantly
faster.
While much of this literature focuses on knowledge
flows as spillovers, some authors have focused on market transactions involving university industry collaboration. Prominent in this stream is the work of Zucker
and Darby (1996, 1998) who examine the role of star
university scientists in the formation and performance
of new firms in biotechnology. Their work points to the
importance of star scientist collaboration in the transfer
of information to nascent firms. Similarly, Thursby and
Thursby (2004) also examine collaboration between scientists and firms but their sample is comprised of firms that license university technologies
and their focus is on contractual mechanism of
transfer.

3. Knowledge ows and incubator rm


performance
In this paper, we build on the ideas from this literature and hypothesize that both contractual and noncontractual mechanisms are important for understanding universityfirm knowledge flows and the effects on
incubator firm performance.
3.1. University licenses and incubator rm
performance
Since the early 1990s, licensing activity in U.S.
research universities has increased considerably. Analyzing the growth of university licensing, Thursby and
Thursby (2002) draw on data from a survey by the Association of University Technology Managers (AUTM)
and state that in 1998 alone, the 132 research universities responding to the survey reported more than 9500
disclosures, more than 4100 new patent applications,
and more than 3000 licenses and options executed.

308

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

The university licensing process starts with a faculty


member disclosing a discovery to the universitys
office of technology transfer. Some universities, like
Georgia Tech, do not wait for faculty to take the first
step of disclosing an invention but rather proactively
monitor university faculty research and encourage
faculty to disclosure inventions. Once a discovery is
disclosed, the office of technology transfer evaluates
the commercial potential of this invention. If there is
some commercial potential and expected licensees are
anticipated, the office of technology licensing applies
for a patent. Note that not all technology licenses go
along with patent protection because many inventions
are protected by copyright, for example, software. Yet,
university technology licenses are generally exclusive.
As a case in point, all of the licenses granted by
Georgia Tech to the incubator firms in this study are
exclusive.
We suggest that exclusive licenses endow the
incubator firm with a unique resource. In a seminal article that laid the theoretical foundation for
the resource-based view of the firm, an important
framework in strategic management research, Barney
(1991, pp. 105106) posited that for firm resource
to have the potential to be the basis of a competitive
advantage, (a) it must be valuable, in the sense that
it exploits opportunities and/or neutralizes threats in
a firms environment, (b) it must be rare among a
firms current and potential competitors, (c) it must be
imperfectly imitable, and (d) there cannot be strategically equivalent substitutes for this resource that are
valuable but are neither rare or imperfectly imitable.
Competitive advantage is defined as a value creating
strategy not simultaneously being implemented by
any current or potential competitors (Barney, 1991,
p. 102).
We thus posit that a technology license fulfills the
attributes discussed by Barney as it is valuable because
it allows the firms to exploit a technological opportunity; it is rare because the license is exclusive and
contains novel technology; it is generally imperfectly
imitable, often protected by legal barriers like patents
or copyrights; and there are generally no readily available substitutes. Thus, holding a technology license
should aid an incubator firm in achieving superior performance because it can implement a strategy based
on this unique resource that its existing or potential
competitors cannot readily imitate.

3.2. Patent citations, absorptive capacity, and


incubator rm performance
Backward patent citations are references made to
prior art in a patent application. Patent backward citations are bibliometric fossils that identify the ideas
on which an incubator firm draws when applying for
a patent. Being able to draw on past research, albeit
public in nature, demonstrates that the incubator firm
is endowed with some degree of absorptive capacity
which enables it to recognize, assimilate, and exploit
external knowledge. Here, it is important to note that
university knowledge, albeit publicly available, is far
from costless. Firms must build internal capabilities
to evaluate external research and apply it to commercial ends (Cohen and Levinthal, 1990). This is often done through hiring intellectual human capital in
the form of star scientists (Zucker and Darby, 1996,
1998), through participation in the broader scientific
community through journal publications (Henderson
and Cockburn, 1994; Cockburn and Henderson, 1998),
and/or through strategic alliances with providers of
the new technology (Rothaermel, 2001). A firms absorptive capacity has been shown to enhance a firms
innovative capability (Cohen and Levinthal, 1989),
which in turn improves firm performance especially
in highly dynamic industries (Rothaermel and Hill,
2005).
We thus suggest that backward patent citations to
university research are indicative of an incubator firms
absorptive capacity to recognize, assimilate, and apply
university knowledge flows to commercial ends. This
is because many capabilities like absorptive capacity
cannot be observed directly. Godfrey and Hill (1995)
argued that unobservable constructs lie at the core of a
number of influential theories in strategic management
research. Given this serious challenge impeding
empirical research, they suggested that what scholars
need to do is to theoretically identify what the
observable consequences of unobservable resources
[capabilities] are likely to be, and then go out and see
whether such predictions have a correspondence in the
empirical world. The analogy here is with quantum
mechanics, which has been confirmed not by observing
subatomic entities (since they are unobservable) but
by observing the trail left by subatomic entities in the
cloud chambers of linear accelerators (Godfrey and
Hill, 1995, p. 530, italics in original).

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

We suggest that a firms absorptive capacity, while


an important construct, is not directly observable. Thus,
we resort to proxying for absorptive capacity by patent
backward citations, which can be understood as indicating the existence of firm-level absorptive capacity
deep within the firm. Moreover, absorptive capacity is
a firm-level capability that is expected to be heterogeneously distributed among firms and thus should lead to
variance in performance. In summary, we suggest that
backward patent citations to university research should
positively enhance incubator firm performance.

4. Methodology
4.1. Research settingGeorgia Techs Advanced
Technology Development Center
The research setting of this study is the Advanced
Technology Development Center, a technology incubator sponsored by the Georgia Institute of Technology. The incubator is located adjacent to the Georgia
Tech main campus in midtown Atlanta as part of a
US$ 250 million state-of-the-art building complex that
houses Georgia Techs Business School and Economic
Development Institute, among others. Besides being
sponsored by Georgia Tech, the ATDC also receives
legislative and financial appropriations from Georgias
Governor and the General Assembly of the state.
The ATDC was founded in 1980 as one of the first
technology incubators in the U.S., and has since generated a cumulative of 4100 jobs and US$ 352 million
in total revenues as of December 31, 1998. During our
study period, the ATDC member firms had a total of
US$ 12 million in annual revenues in 1998, US$ 19
million in 1999, and US$ 18 million in 2000. In the
late 1990s, Georgia Techs ATDC was voted as one of
the top incubators in the U.S. based on a survey of peer
incubators conducted by Inc. magazine (Rosenwein,
2000). The ATDC focuses on incubating early stage
companies (03 years), with the companys founding
date generally coinciding with the firms admission to
membership into the incubator.
The ATDC managers actively solicit applications
from new ventures, and admitted, during our study period, between 10 and 20% of their applicants after a
fairly stringent, two-staged review process. It is not
necessary that the technology underlying the new ven-

309

ture is related to Georgia Tech; yet, it must be proprietary in nature. During the last few years, the size of
the full-time professional staff of the ATDC remained,
despite turnover, fairly constant at 22 managers. These
managers assist the commercialization efforts of the
ATDC member firms.
4.2. Sample and data
The sample consists of the population of member
firms in the ATDC for the years 19982000. A total
of 79 firms were tenants of the ATDC during this 3year time frame. The year 1998 marks the first year
detailed data were collected for the firms in the incubator. We drew our sample based on the years 19982000
to be able to follow each firm for a minimum of 4 years
to assess the performance of the incubator firms. Employing multiple performance measures, we assessed
the performance of the newly formed ventures over or
at the time period t + 1, where t 3 years. This time
window appears to be a conservative one given the fact
that incubator tenants tend to graduate from public incubators within 2 years and from private incubators
within 1 year (Rosenwein, 2000). While the ATDC has
no explicit graduation policy, it attempts to graduate
their members in a timely fashion. In the year t + 1,
the technology venture could fall into one of three categories: (1) failure, i.e., the firm ceased to exist due
to bankruptcy or liquidation; (2) firm remains in the
incubator; and (3) successful graduation, i.e., the firm
is a stand-alone going concern or was acquired. We
included acquisitions as part of successful graduation
based on qualitative assessments made by ATDC managers.
Data for the 79 firms were collected annually for the
6-year time period between 1998 and 2003 through a
survey instrument that was administered to all firms in
the sample in the spring of every year to collect data
for the prior year. Accordingly, data collection began
in the spring of 1999 and ended in the spring of 2004.
This longitudinal, repeat survey approach allowed us
to obtain multiple, ubiquitous performance outcomes
for all 79 firms in the initial sample. Thus, our results
are not prone to a survivor bias, frequently observed
in studies focusing on new venture creation and their
early performance.
For the subset of firms based on Georgia Tech technologies, the Georgia Institute of Technologys Of-

310

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

fice of Technology Licensing provided us with data


on relevant patents and founding dates. We augmented
the collection of the quantitative data through semistructured interviews with managers of the ATDC, the
Institutes Vice Provost for Economic Development
and Technology Ventures, the Institutes Director of
the Technology Licensing Office, and the Institutes
Director of its VentureLab, a center founded to identify commercializable technologies within the Institute.
A third source of data was the patent database maintained by the U.S. Patent and Trademark Office, an
agency of the U.S. Department of Commerce. Here,
we accessed all patents awarded to the incubator firms
in this sample.
4.3. Measures
4.3.1. Incubator rm performance
Incubator firm performance is the dependent variable of this study. Clearly, assessing the performance of
entrepreneurial start-ups, and incubator firms in particular, is a thorny problem retarding empirical research
in this important area (Phan et al., 2004). Based on
the annual repeat survey instrument underlying the
data collection for this study, we are fortunate to assess the performance of incubator firms on multiple
dimensions including revenues, total funds raised, venture capital funding obtained, and whether the firm
graduated, failed, or remained in the incubator. Assessing the performance of incubator firms along several performance dimensions is particularly salient in
this context because the most appropriate performance
metrics for nascent technology ventures are less than
clear.
4.3.1.1. Revenues. One of the performance metrics we
employed is total cumulative revenues obtained by the
incubator firms. To enhance the validity of this measure, we did assess it as cumulative revenues accrued
over the time period including t + 1 to avoid dependence
on single observations often characterized by high annual fluctuations. While revenues are an accepted performance metric for more mature firms, it is less clear if
this measure is suitable for the incubator context. We attempt to shed some more light on this issue. We applied
a logarithmic transformation to enhance the normality
of this variable.

4.3.1.2. Total funds raised. A second performance


metric used is the total amount of cumulative funding
the new ventures obtained over the time period including t + 1. We constructed the total funds raised variable
by leveraging fine-grained data pertaining to the different financing sources: family and friends, angel investors, venture capitalists, private placements, equity
investments, and grants.
4.3.1.3. VC funding. One important milestone in the
development of a nascent technology venture is obtaining venture capital funding (Shane and Cable, 2002;
Shane and Stuart, 2002). Funding obtained from
venture capitalists takes on an important signaling role
as it often bestows legitimacy upon the new venture
(Stuart et al., 1999). Moreover, some universities,
albeit not the focal institution of this study, make
obtaining a university license contingent upon having
received venture capital. We assessed whether the
incubator firms in this sample have obtained venture
capital during the time period including t + 1 by a bivariate indicator variable taking on the value of 1 if the
incubator firm received venture capital funding, and 0
otherwise.
4.3.1.4. Failure, graduation, and remain in incubator. As discussed in detail in Rothaermel and Thursby
(2005), one of the important milestones in the development of incubator firms is the timely graduation from
the incubator. On an average, private incubators expect
their tenants to graduate within 1 year, while public incubators expect their tenants to graduate within 2 years
(Rosenwein, 2000). To be conservative, we assessed
the state of incubator firms in t + 1, where t 3 years.
In the year t + 1, an incubator venture could fall into one
of three categories: (1) failure, i.e., the firm ceased to
exist due to bankruptcy or liquidation; (2) firm remains
in the incubator; and (3) successful graduation, i.e., the
firm is a stand-alone going concern or was acquired.
We subsumed acquisitions under successful graduation
because the few cases in which incubator firms were
acquired in this sample (three firms or 4%) reflect successes rather than failures based on the evaluations by
ATDC managers. We coded the performance of the new
technology ventures in t + 1 as a multinomial variable
with three categories: failure, graduation, and remaining in incubator. Remaining in the incubator serves as
reference category.

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

4.3.2. Knowledge ows from university to


incubator rms
The key independent construct of this study concerns knowledge flows from the university to the incubator venture. Here, we hypothesized that exclusive
knowledge flows in terms of a university license can endow the start-up with a unique resource that should lead
to a superior performance (Barney, 1991). Moreover,
we suggested that university backward patent citations
are indicative of a start-ups absorptive capacity that
enables it to recognize public knowledge spillovers emanating from the university, assimilate them internally,
and then to apply them to commercial ends (Cohen
and Levinthal, 1989). To obtain a comprehensive and
fine-grained assessment of knowledge flows from the
university to incubator firms, we employed five distinct
variables proxying for different mechanisms through
which knowledge may flow from the university to an
incubator firm. In particular, we proxied for knowledge
flows from the sponsoring university as well as more
general university knowledge flows emanating from
the broader university community.
4.3.2.1. GT license. One mechanism through which
knowledge can flow from a university to an incubator
firm is through a licensing agreement. Here, we assessed potential knowledge flows from the sponsoring
university, Georgia Tech, to the incubator firms by including a variable that tracks whether the firm in the
sample was founded to commercialize a technology
invented at Georgia Tech and subsequently licensed
it from the Institutes Office of Technology Licensing
(1 = GT license). These licenses are exclusive in the
sense that they are only given to one firm.
4.3.2.2. Backward citations to university research. A
second area where knowledge flows from universities
to incubator firms should manifest themselves is in
the incubator firms patent citations because all prior
art must be credited in the patent application. Patents
reflect inventions because they are only granted to
processes or products that are novel, non-obvious, and
industrially useful as judged by someone possessing
proficient knowledge in the relevant technical area (Acs
and Audretsch, 1989).
Here, assessing knowledge flows from a university
to an incubator firm can be accomplished by analyzing the backward citations to university research in a

311

technology ventures patent portfolio. To do this, we


secured copies of all patents that the start-ups in the
sample had obtained. We then counted all backward
citations to university research in an incubator firms
patent portfolio. University research is defined as either
citations to patents granted to universities or citations to
academic journals publishing research results. In the final step, we took the ratio of an incubator firms patent
backward citations to university research over its total number of patent backward citations to assess the
magnitude to which the new venture is drawing on university research in their own inventions. This measure
can be considered as a proxy for knowledge flows from
universities to incubator start-ups.
4.3.2.3. Backward citations to academic journals. We
suggest that research findings published in academic
journals tend to be more embryonic and basic in nature
than research that is explicated in university patents,
which tend to be more developed and explicit. In general, university faculty tend to first publish research results in academic outlets prior to the university applying
for a patent. For example, in 1973, Stanley Cohen (then
a professor at Stanford) and Herbert Boyer (then a professor at the University of California, San Francisco)
first published their scientific breakthrough in recombinant DNA in the Proceedings of the National Academy
of Sciences of the United States of America (Cohen
et al., 1973). The patent on recombinant DNA, however, was granted 7 years later in 1980, and assigned
to Stanford University with Cohen and Boyer listed as
inventors (U.S. Patent 4,237,224). Therefore, to assess
the potential flow of early stage, basic knowledge to
incubator firms, we included a ratio of an incubator
firms patent backward citations to academic journals
over its total number of patent backward citations.
4.3.2.4. Backward citations to GT research. Besides
highlighting a technology license of the sponsoring
university as one possible mechanism through which
knowledge from the sponsoring university can flow
to an incubator firm, we also assessed the impact of
knowledge flows from the sponsoring university by including a ratio of the firms patent backward citations
to Georgia Tech research over its total number of patent
backward citations. This measure indicates how much
the incubator firm draws on localized knowledge. Georgia Tech research is defined as either citations to patents

312

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

granted to Georgia Tech or citations to research published by Georgia Tech faculty members.
4.3.2.5. Backward citations to non-GT research. Besides focusing on knowledge flows from the sponsoring
university, we also consider the impact of knowledge
flows from research universities that are not directly
linked to the focal incubator under consideration. Here,
we assess the impact of the ratio of firms patent backward citations to non-GT research over its total number of patent backward citations. Non-GT research is
defined as either citations to patents granted to any university other than GT or citations to research published
by any person that is not a GT faculty member. Please
note that the sum of backward citations to GT research
and backward citations to non-GT research equates to
Backward Citations to University Research, the first
backward citation measure introduced above.
4.3.3. Control variables
We included a number of control variables that theoretically could impact new venture performance.
4.3.3.1. Firm size. When assessing the performance
of incubator firms, it is critical to control for their firm
size. Because the important assets of incubator firms
tend to be intangible in nature, it is more appropriate to
use the number of employees as a proxy for firm size
(employees) as done in prior research focusing on hightechnology ventures (Rothaermel, 2002; Rothaermel
and Deeds, 2004).
We controlled for firm size effects through the number of employees up to the year prior to which the outcome variable was assessed. Moreover, because newly
created ventures tend to be quite small, we collected
data not only on the number of full-time employees, but
also on the number of part-time employees. Each fulltime employee was counted as one employee, while
one part-time employee was counted as one-half of a
full-time employee.
4.3.3.2. Industry effects. When assessing new venture
performance, it is pertinent to control for industry effects. We tracked each incubator firms industry based
on their Standard Industry Classification (SIC) codes.
About 60% of the firms were active either in the software industry or the in telecommunications industry. To

control for these two most prevalent industries, we inserted two indicator variables in the regression models.
The first indicator variable takes on 1 if the incubator
firm is a software company, and 0 otherwise. The second indicator variable takes on 1 if the incubator firm
is a telecom company, and 0 otherwise.
4.3.3.3. Time in incubator. While the sample is not
prone to a survivor bias, we are faced with the problem
of left censoring because the ATDC technology incubator was in existence prior to 1998, the first year of
our annual data collections. To ameliorate this problem, we recorded the year that each firm was admitted
into the incubator, which generally coincides with the
firms founding date, and the last year the firm remained
in the incubator. These two data points enabled us to
construct the time in incubator variable, which is the
number of years the firm remained in the technology
incubator, to account for left censoring.
4.3.3.4. Non-GT university link. When assessing the
effect of university knowledge flows on incubator firm
performance, it is prudent to control for university linkages that the ATDC ventures may have to other, nonsponsoring universities. In fact, the sample firms listed
linkages to 11 other U.S. research universities besides
Georgia Tech. To isolate the effect of different knowledge flow mechanisms on the performance of ATDC
ventures, we created an indicator variable that takes on
the value of 1 if the firm had a link to a university other
than Georgia Tech. Some ATDC firms in the sample,
for instance, maintained a link to a university other than
Georgia Tech but did not have a link to Georgia Tech.
4.3.4. Estimation procedures
When assessing the performance of incubator firms,
we focus on four different outcome variables: revenues,
total funds raised, venture capital obtained, and graduation, failure, or remaining in the incubator. These
different dependent variables indicate different estimation procedures. The regression models with revenues
and total funds obtained as dependent variables were
estimated using ordinary least squares (OLS).
Venture capital obtained is a binary variable taking on 1 if the incubator firm received venture capital,
and 0 otherwise. This model indicates logit regression.
 , is the probability of the venThe outcome variable, Y
ture receiving or not receiving venture capital based

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

on a non-linear function with two outcomes. The logit


model is estimated with a maximum likelihood procedure and has the following specification:




Y
ln
=+
j Xij ,

1Y
where Xij is a vector of independent variables.
The last performance variable employed in this
study can take on three categories: failure, remaining
in incubator, and successful graduation. This indicated
application of a multinomial logistic regression, estimated with a maximum likelihood procedure. The outcome variable, Pj , is the probability of falling into one
of the outcome categories based on a non-linear function with three outcomes (Maddala, 1983):
ej x
Pj =
(j = 1, 2, . . . , m 1)
D
and
1
Pm =
D
where
D=1+

m1


ej x .

k=1

5. Results
We assessed the effect of universityincubator firm
knowledge flows on 79 firms incubated in the Georgia Techs ATDC over the 6-year time period between
1998 and 2003. We employed multiple performance
measures to reflect the multi-dimensional nature of incubator firm performance. In particular, we assessed
the performance of the newly formed ventures over or
at the time period t + 1, where t 3 years. Relying on a
repeat sample method, we were able to obtain ubiquitous outcome variables on all 79 firms, thus overcoming
a potential survivor bias.
We tracked incubator firm performance on four different dimensions. In total, the 79 incubator firms raised
over US$ 404 million in funding, with a US$ 193 million (48%) thereof being VC funding. The average incubator firm had revenues of US$ 1.6 million and had
raised a total of US$ 5.1 million in funding. Notable is
also the high variance in these performance variables.

313

While some incubator firms did not generate any revenues or raised any funding, one firm accrued over US$
39 million in revenues and another firm raised a total
of US$ 36.5 million in funding. Almost one-half of the
sample incubator firms (46%) obtained VC funding. In
the year t + 1, 23 firms (29%) had graduated successfully, 15 firms (19%) remained in the incubator, while
41 firms (52%) had failed.
The key independent variables of this study proxy
for universityincubator firm knowledge flows. We find
that 11 firms (14%) were founded based on a Georgia Tech license. Of the 79 incubator firms, 13 firms
(16.5%) had rights (either through a Georgia Tech license or as an assignee themselves) to a total of 35
patents from the U.S. Patent and Trademark Office. It is
important to note that the firms founded on GT licenses
in comparison to non-GT start-ups that were assigned
patents did not differ with respect to the total number
of patents granted or any of the four different patent
backward citations measures used in this study. This
implies that the analysis of universityincubator firm
knowledge flows based on patent citation measures is
unbiased with respect to the firm having obtained a GT
license or not.
Overall, these 35 patents contained 978 citations,
splitting into 766 backward citations and 212 forward citations.1 The 766 backward citations split into
627 non-university (i.e., industry) backward citations
(82%) and 139 (18%) university backward citations. Of
the latter, 22 (16%) were backward citations to Georgia
Tech research.
At the firm level, about 3.3% of all patent citations
referenced university research in general, while 2.4%
of all patent citations were traced back to academic
journals in particular. When dividing the university
backward citations in citations to the sponsoring university and citations to other non-sponsoring university
research, we find that 0.7% of all backward citations
are to Georgia Tech research, while 2.6% of all
backward citations are to non-GT university research.
The average patent citations to university research are
1 When analyzing the backward citations, we also coded whether
the citation was added by the inventor or by the examiner. Since 2001,
patent citations added by the examiner are marked by an asterisk. In
this sample, there were 397 backward citations since 2001. We found
that only six of these citations (0.15%) were added by the patent
examiner. We are thus confident that the results are not materially
influenced by patent citations added by the patent examiner.

314

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

quite low because most incubator firms (66 firms or


83.5%) did not obtain any patents. Thus, our econometric estimates for university patent citation ratios
on incubator firm performance are quite conservative,
and potentially biased downward. When assessing
the prevalence of university patent backward citations
among the firms that have been granted patents, we
found that 18% of all their patent citations was to university research, which split into 14% to non-Georgia
Tech research and 4% GT research. When considering
the more narrow set of academic journal citations, we
find that among these firms, 12% of all patent citations
are to academic journals. Indeed, some of the firms exhibited very high university backward citations in their
patent portfolios. For example, the maximum ratio
for overall university backward citations was 83%, for
backward citations to academic journals it was 79%,
and for backward citations to Georgia Tech research it
was 32%.
Noteworthy is also the discriminant validity of
the different knowledge flow measures employed in
this study. When excluding bivariate correlations that
share by definition a significant amount of common
variance,2 and which are not inserted in the same regression models, we find that the bivariate correlations among the different knowledge flow measures
are well below the suggested cut-off point of R = 0.70,
suggesting satisfactory discriminant validity (Cohen et
al., 2003).
The control variables reveal that the average incubator firm had about 14 employees and spent 2.4 years in
the ATDC. The majority of firms are either active in the
software industry (34 firm or 43%) or in the telecommunications industry (13 firms or 17%).3 A little more
2 That is, backward citations to university research and backward
citations to academic journals; backward citations to university research and backward citations to non-GT research; backward citations to academic journals and backward citations to non-GT research.
3 In total, the 79 firms fall into 14 different industries based on fourdigit SIC codes. Besides software and telecommunications, there
are six firms each in manufacturing and communications (8%), five
in healthcare (6%), three Internet businesses (4%), two firms each
in agriculture, biotechnology, environmental services, and general
services (3% each), and one firm each in microelectronics, paper
industry, robotics, and video industry (1% each). We suggest that
the results presented below appear not to be directly influenced by
the Internet boom and bust during the late 1990s and early 2000s
because only three firms in the sample are Internet firms.

than 20% of the firms had a linkage to a research university other than Georgia Tech. Table 1 depicts the
descriptive statistics and bivariate correlation matrix,
while Table 2 presents the regression results.
We advanced two exploratory hypotheses. We argued that exclusive knowledge flows in terms of a university license can endow the start-up with a unique
resource, which can lead to a competitive advantage
(Barney, 1991). We also suggested that university backward patent citations are indicative of a start-ups absorptive capacity that enables it to recognize public
knowledge flows emanating from a university, assimilate them internally, and then to apply them to commercial ends (Cohen and Levinthal, 1989). A new ventures
absorptive capacity is hypothesized to positively affect
its performance.
Each of the four different performance measures
was assessed in three different regression models. The
four different performance dimensions are revenues,
funds obtained, venture capital, and graduation from
the technology incubator. In the first model of each
three-model block, we added the effect of backward citations to university research on incubator firm performance. In the second model, we evaluated the impact of
a somewhat more stringent knowledge flow measure,
backward citations to academic journals, on incubator firm performance. In the last model of each block,
we split the backward citations to university research
into backward citations to research by the sponsoring
university, Georgia Tech, and backward citations to research by other universities. Each regression model,
however, contains the proxy for a GT license.
Models 13 evaluate the effect of the five different
knowledge flow mechanisms on incubator firm performance proxied by revenues. Here, we find that none
of the knowledge flow proxies reach significance. This
might be indicative of the fact that revenues is not the
most appropriate measure when assessing the performance of nascent technology ventures.
Models 46 assess the impact of the different knowledge mechanisms on the total amount of funds raised
by the incubator firms. All three models are overall
highly significant (p < 0.001), and exhibit R2 values of
around 0.36. With respect to individual coefficients,
the regression results reveal that backward citations to
university research (Model 4), backward citations to
academic research (Model 5), and backward citations
to non-GT research (Model 6) are each positive and

1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.

Revenues
Total funds raised
VC funding
Failure
Graduation
Remain in incubator
Employees
Software
Telecom
Time in incubator
Non-GT university link
GT license
Backward citations to
university research
Backward citations to
academic journals
Backward citations to
GT research
Backward citations to
non-GT research

Mean

S.D.

1,654,023
5,116,468
0.456
0.519
0.291
0.190
13.734
0.430
0.165
2.430
0.203
0.139
0.033

5,338,124
7,779,855
0.05
0.501
0.01
0.32
0.503
0.29 0.28 0.19
0.457
0.25
0.35
0.14 0.67
0.395
0.08 0.05
0.08 0.50 0.31
16.152
0.23
0.50
0.41 0.07
0.30 0.26
0.498
0.03
0.24
0.18
0.02
0.06 0.09
0.17
0.373
0.07
0.05
0.21 0.05
0.09 0.04
0.25
2.164
0.24 0.33 0.23 0.03 0.28
0.37 0.25
0.404
0.09 0.14 0.02
0.17 0.32
0.16 0.14
0.348
0.05 0.01
0.00 0.20 0.10
0.36 0.14
0.123
0.03
0.00
0.05 0.25 0.07
0.40 0.11

0.024

0.101

0.007

0.038

0.026

0.107

0.09

10

11

12

0.39
0.19 0.07
0.13 0.22 0.02
0.28 0.02 0.26
0.22 0.12 0.24

0.02
0.30 0.38

13

14

0.09 0.19 0.11

0.37 0.10

0.18 0.10 0.17

0.31 0.24

0.95

0.10 0.08 0.09 0.19 0.07

0.32 0.10

0.16 0.08 0.28

0.24 0.36

0.56

0.44

0.35 0.09

0.20 0.11 0.17

0.26 0.32

0.96

0.94

0.07

0.01

0.03

0.09 0.22 0.06

15

0.29

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

Table 1
Descriptive statistics and bivariate correlation matrix

N = 79.

315

316

Table 2
Regression results assessing the impact of universityincubator firm knowledge flows on incubator firm performance

R2
F-stat
2 log likelihood

Constant
Employees
Software
Telecom
Time in incubator
Non-GT university link
GT license
Backward citations to university research
Backward citations to academic journals
Backward citations to GT research
Backward citations to non-GT research
R2
2 log likelihood

Standard errors in parentheses.


p < 0.10.
p < 0.05.
p < 0.01.
p < 0.001.

Model 2
Revenues
3.683**** (0.299)
0.803*** (0.334)
0.142 (0.365)
0.075 (0.356)
0.939*** (0.324)
0.061 (0.329)
0.111 (0.325)

Model 3
Revenues
3.683**** (0.301)
0.794*** (0.338)
0.178 (0.370)
0.090 (0.360)
0.885*** (0.331)
0.151 (0.336)
0.056 (0.348)

0.315 (0.336)
0.233 (0.348)
0.260 (0.344)
0.159
1.921*

0.167
2.035*

Model 4
Model 5
Model 6
Model 7
Funds raised
Funds raised
Funds raised
VC funding
****
****
****
5.11E06
(7.36E05)
5.11E06
(7.36E05)
5.11E06
(7.42E05) 0.184 (0.278)
****
****
****
0.410
(0.106)
0.414
(0.106)
0.411
(0.107)
0.919*** (0.396)
0.210** (0.116)
0.206** (0.116)
0.209** (0.117)
0.638** (0.329)
0.007 (0.115)
0.014 (0.113)
0.007 (0.109)
0.610** (0.318)
0.247*** (0.103)
0.243*** (0.102)
0.243** (0.105)
0.469* (0.334)
0.152* (0.105)
0.152* (0.104)
0.148 (0.106)
0.005 (0.293)
0.109 (0.108)
0.131 (0.103)
0.114 (0.110)
0.230 (0.293)
0.155* (0.114)
0.441* (0.306)
*
0.150 (0.107)
0.019 (0.110)
0.146* (0.109)

0.168
1.767*

0.355
5.586****

0.356
5.616****

0.356
4.833****

Model 8
VC funding
0.183 (0.280)
0.934*** (0.398)
0.644** (0.326)
0.600** (0.317)
0.417* (0.336)
0.027 (0.294)
0.291 (0.290)
0.500* (0.340)

0.353

0.363

84.665****
Model 11

83.859****

Model 9

Model 10

VC funding
0.193 (0.282)
0.920*** (0.394)
0.629** (0.328)
0.603** (0.318)
0.420 (0.333)
0.031 (0.297)
0.275 (0.302)

Fail

Graduate

Fail

Graduate

Model 12
Fail

Graduate

1.097* (0.676)
0.612* (0.419)
0.483 (0.394)
0.300 (0.315)
0.079 (0.347)
0.334 (0.365)
5.620**** (1.496)

1.346** (0.686)
0.334 (0.486)
0.357 (0.439)
1.272*** (0.502)
2.039*** (0.860)
0.514 (0.421)
0.507 (0.481)

0.749* (0.553)
0.466 (0.385)
0.349 (0.372)
0.344 (0.290)
0.136 (0.332)
0.373 (0.360)

0.021** (0.564)
0.268 (0.439)
0.316 (0.408)
1.147*** (0.480)
1.350** (0.643)
0.255 (0.383)

3.335** (1.593)
0.949** (0.533)
0.651* (0.468)
0.469 (0.379)
0.023 (0.408)
0.929** (0.494)

3.610** (1.597)
0.667 (0.603)
0.548 (0.517)
1.608*** (0.583)
5.380*** (2.162)
1.069** (0.624)

4.850**** (1.534)
0.207 (0.542)
0.509* (0.334)
0.363
83.837****

0.147 (0.950)
18.036*** (6.244)
0.775 (1.234)

0.546
119.663****

0.468
129.753****

0.654
103.452****

0.067 (2.077)
1.375 (1.175)

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

Constant
Employees
Software
Telecom
Time in incubator
Non-GT university link
GT license
Backward citations to university research
Backward citations to academic journals
Backward citations to GT research
Backward citations to non-GT research

Model 1
Revenues
3.683**** (0.300)
0.803*** (0.337)
0.171 (0.369)
0.086 (0.360)
0.930*** (0.326)
0.112 (0.332)
0.108 (0.342)
0.163 (0.362)

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

significant (p < 0.10) in predicting incubator firm performance.


The results of the logit estimations, predicting the
probability of obtaining VC funding, are displayed in
Models 79. The results for obtaining VC funding are
basically identical to the results when applying total
funds raised as dependent variable. All three models
are overall highly significant (p < 0.001), and exhibit R2
values of around 0.36. Here again, backward citations
to university research (exp() = 1.554, Model 7), backward citations to academic research (exp() = 1.649,
Model 8), and backward citations to non-GT research
(exp() = 1.663, Model 9) are each positive and significant (p < 0.10) in predicting incubator firm performance.
Models 1012 display the results when assessing the
impact of the different knowledge flows on the incubator firms probability of failure and graduation, while
remaining in the incubator serves as reference category. Each of the three models is statistically significant
(p < 0.01 or smaller), and the R2 ranges between 0.47
and 0.65. Holding a GT license reduces the probability
of outright failure (Model 12: exp() = 0.395, p < 0.05),
but also retards timely graduation from the incubator
(Model 12: exp() = 0.343, p < 0.05). With respect to
backward patent citations, we find that backward citations to university research (Model 10: exp() = 0.004,
p < 0.001), backward citations to academic journals
(Model 11: exp() = 0.008), and backward citations to
GT research (Model 12: exp() = 1.47E8, p < 0.01),
each significantly reduces the probability of outright
failure.
Some of the results for the control variables are
also noteworthy. Firms that grow faster in terms of
employees perform significantly better along all the
different performance metrics used to assess incubator
firm performance in this study. Yet, these faster growing firms are also somewhat more likely to fail. The
results also manifest consistent industry effects. Software firms tend to perform significantly higher when
considering the amount of funding raised, the probability of obtaining VC funding, and exhibit a lower
likelihood of outright failure. Telecom firms are significantly more likely to obtain VC funding and less likely
to experience failure. Firms that remain longer in the
incubator tend to raise fewer funds, are less likely to
obtain VC funding, and are less likely to graduate in
a timely fashion. Firms that stay longer in the incuba-

317

tor, however, tend to generate significantly higher revenues. Incubator firms that maintain linkages to other
research universities than Georgia Tech tend to raise
significantly fewer funds and are less likely to graduate within 3 years or less.
5.1. Robustness checks
We checked if multi-collinearity could bias the results when applying OLS estimations (Models 16).
Here, we found that the maximum variance inflation
factor was 1.5, well below the suggested cut-off point
of 10 (Cohen et al., 2003). Multi-collinearity, however,
did appear to affect the results when including an additional control variable for an incubator firms total number of patents received or a binary indicator variable
whether the firm received a patent or not. The results for
the probability of obtaining VC funding and the results
with respect to failure or graduation remained robust;
however, the results for predicting the total amount of
funding raised do not reach significance. The overall
somewhat weaker results can be explained by the fact
that the variable total number of patents received and
the indicator variable patent received (=1) are highly
significantly correlated with each of the four backward
patent citation measures employed in this study (at
p < 0.01 or smaller). This high correlation is expected
because the firms that obtain patents are the only ones
that can cite university research in their patents. Therefore, it appears prudent to not include the patent count
or indicator variables thereof simultaneously with the
backward patent citations measures in the regression
estimations.

6. Discussion
One of the arguments for incubators associated with
universities is that knowledge flows from universities
should enhance performance of high-technology ventures and that access to this knowledge is not free,
despite the publication norms of science. In this paper, we examined two mechanisms by which incubator
firms can access this knowledge. One, which is available to new ventures based on Georgia Tech inventions, is a license to develop and use a university invention. In the case of ATDC firms, all of the licenses
to Georgia Tech inventions were exclusive so that the

318

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

resource-based view of the firm suggested that these


licenses provide a unique, performance-enhancing resource (Barney, 1991). The other mechanism we explored was citations to university research found in the
patents associated with incubator firms. We argued that
this mechanism reflects not only a non-exclusive means
for firms to access university knowledge, but also that
it reflects the ability of the new venture to utilize university knowledge, and thus could be seen as a proxy
for absorptive capacity (Cohen and Levinthal, 1990).
We argued that both licenses and backward citations
should be positively related to incubator firm performance. A secondary purpose of this study was to examine different performance metrics for evaluating the
performance of nascent technology ventures.
We found little support for the unique resource hypothesis because the license variable was in general not
significant. Holding a Georgia Tech license had a significant effect only on performance measured by failure
and graduation, and then only when the citation ratios
were split between Georgia Tech and non-Georgia Tech
citations. This result is interesting because this is also
the only model for which the localized citation variable, the ratio of citations to Georgia Tech research to
all citations, is significant.
We found more consistent support for the absorptive
capacity hypothesis as backward citations to university research positively affect three of the performance
measures. This was the case both for all university citations and citations to academic publications alone.
Notice that the variable backward citations to university research is, ceteris paribus, one minus the ratio of
citations to industry patents to all citations (abstracting
from non-university and non-industry citations). Thus,
the coefficient for the university research variable is
the opposite of the coefficient had we entered the ratio of industry citations. A natural interpretation of the
university citation variable is that it is a measure, not
only of absorptive capacity, but also of how basic are
the patents of the new venture. According to this interpretation, our results of citations on the probability of
obtaining VC funding suggest that venture capitalists
tend to be more likely to back basic inventions rather
than those whose prior art arises primarily in industry.
With respect to evaluating different performance
metrics for technology ventures, in particular in the
context of university knowledge flows, we found that
revenues were a poor measure for incubator firm perfor-

mance. The revenue models explain the lowest amount


of variance, and none of the individual proxies for
knowledge flows reaches statistical significance. It is
not surprising that revenues appear to be an inappropriate measure for assessing the performance of incubator firms because the firms are very young (less than
3 years old) and compete in the high-technology space,
where many firms do not generate much revenues initially as they invest to develop the new technology.
More promising performance metrics appear to be the
total amount of funding obtained, whether the firm was
backed by VC funding, and whether the firm graduated
from the incubator in a timely manner or failed altogether. The last variable should be applied with caution
in future research because many incubators tend to have
an explicit graduation policy (not the ATDC) and encourage or expect timely graduation. Thus, total funds
obtained or VC funding are market mechanisms that
appear to assess the performance of new technology
ventures in a satisfactory manner.
6.1. Limitations and future research
One of our more important assumptions underlying
this research was that patent citations reflect knowledge flows. While this assumption is based on a long
tradition in economics, some recent research has suggested that patent citations may not reflect knowledge
flows, since citations are often added by attorneys and
patent examiners (Jaffe et al., 2000; Alcacer and Gittelman, 2004; Sampat, 2004). While this is a fundamental
challenge to the research relying on patent data when
attempting to capture knowledge flows, the recent practice of publishing examiner cites mitigates this problem. Moreover, if patent citations indeed do not truly
reflect knowledge flows, future research needs to develop alternate metrics that capture knowledge flows
more effectively.
We suggested that firms with a higher ratio of university citations in their patent portfolio achieve higher
performance because these firms are based on more
basic invention with a greater potential of making a
commercial breakthrough. While the ratio of university citations to total citations was a positive predictor
of incubator firm performance, especially with respect
to total amount of funding obtained and the probability of obtaining VC funding, we need to emphasize
that the relationship between the ratio of university ci-

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

tations to total patent citations on firm performance


may not be positive and linear because the average
firm in this sample exhibited a low level of university citations among its patents. Future research could
investigate a potentially diminishing marginal or even
diminishing total returns hypothesis for the relationship between university backward citations and firm
performance.
Clearly, future research should also address the generalizability of our findings. While we find some support for the absorptive capacity hypothesis, we need
to emphasize that our research setting is somewhat
unique. While it enabled us to empirically assess interfirm performance differentials among incubator firms,
we were limited to firms incubated in the ATDC sponsored by Georgia Tech, a research institute with a clear
focus on engineering sciences. Some researchers have
begun to compare incubator differential performance
(Mian, 1996; Colombo and Delmastro, 2002); however,
studies on interfirm differential performance of incubator firms are rare. We hope that future research will
be able to apply a repeat survey approach similar to the
one used in this study in order to collect fine-grained,
longitudinal data on incubator firms across different
technology incubators.
6.2. Implications for public policy
The results presented in this study seem to indicate
that knowledge does flow, via different mechanisms,
from universities to incubator start-up firms. While the
negative impact of having a link to a research university other than the sponsoring university on firm performance seems to indicate some benefits of being closely
located to the sponsoring university, other measures for
localized spillovers did generally not reveal significant
results. One area of public policy concern could therefore be the enhancement of localized spillovers because
many public incubators and other university-based
technology initiatives are formed to improve the economic performance of the region. Siegel et al. (2003),
for example, showed, when examining the effect of university science parks in the UK on firm research productivity, that firms associated with science parks were
more productive than those not so located. The question
of how localized spillovers from universities to incubator firms can be enhanced is an interesting one and
opens up another promising avenue for future research.

319

Acknowledgements
We thank Tony Antoniades (of the ATDC), George
Harker (Director, Office of Technology Licensing,
Georgia Institute of Technology), and H. Wayne
Hodges (Vice Provost for Economic Development and
Technology Ventures, Georgia Institute of Technology)
for their generous support and invaluable input, Stuart Graham for comments and suggestions, and Shanti
Dewi for research assistance. A prior version of this
paper was presented at the 2004 Technology Transfer
Society (T2S) Conference. We thank the special issue
editors and the session attendants for valuable input.
All remaining errors and omissions are entirely our
own. Thursby gratefully acknowledges research support from NSF SES 0094573.
References
Acs, Z., Audretsch, D.B., 1989. Patents as a measure of innovative
activity. Kyklos 42, 171180.
Adams, J., 1990. Fundamental stocks of knowledge and productivity
growth. Journal of Political Economy 98, 673702.
Agrawal, A., Henderson, R., 2002. Putting patents in context: exploring knowledge transfer from MIT. Management Science 48,
4460.
Alcacer, J., Gittelman, M., 2004. How do I know what you know?
Patent examiners and the generation of patent citations. Working
Paper. New York University.
Barney, J., 1991. Firm resources and sustained competitive advantage. Journal of Management 17, 99120.
Branstetter, L., 2004. Is academic science driving a surge in industrial innovation? Evidence from patent citations. Working Paper.
Columbia University.
Cockburn, I.M., Henderson, R.M., 1998. Absorptive capacity, coauthoring behavior, and the organization of research in drug discovery. Journal of Industrial Economics 46, 157182.
Cohen, S.N., Chang, A.C.Y., Boyer, H.W., Helling, R., 1973. Construction of biologically functional bacterial plasmids in vitro.
Proceedings of the National Academy of Sciences of the United
States of America 70, 32403244.
Cohen, J., Cohen, P., West, S.G., Aiken, L.S., 2003. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences.
Erlbaum, Mahwah, NJ.
Cohen, W.M., Levinthal, D.A., 1989. Innovation and learning: the
two faces of R&D. Economic Journal 99, 569596.
Cohen, W.M., Levinthal, D.A., 1990. Absorptive capacity: a new
perspective on learning and innovation. Administrative Science
Quarterly 35, 128152.
Cohen, W.M., Florida, R., Randazzes, L., Walsh, J., 1998. Industry
and the academy: uneasy partners in the cause of technological
advance. In: Noll, R. (Ed.), Challenges to Research Universities.
The Brookings Institution, Washington, DC, pp. 171200.

320

F.T. Rothaermel, M. Thursby / Research Policy 34 (2005) 305320

Colombo, M.G., Delmastro, M., 2002. How effective are technology incubators? Evidence from Italy. Research Policy 31, 1103
1122.
Griliches, Z., 1992. The search for R&D spillovers. Scandinavian
Journal of Economic 94, 2947.
Godfrey, P.C., Hill, C.W.L., 1995. The problem of unobservables in
strategic management research. Strategic Management Journal
16, 519533.
Henderson, R., Cockburn, I., 1994. Measuring competence? Exploring firm effects in pharmaceutical research. Strategic Management Journal 15, 6384 (winter special issue).
Henderson, R., Jaffe, A., Trajtenberg, M., 1998. Universities as a
source of commercial technology: a detailed analysis of university patenting. Review of Economics and Statistics 80, 119127.
Jaffe, A.B., 1989. Real effects of academic research. American Economic Review 79, 957970.
Jaffe, A.B., Trajtenberg, M., Fogarty, M., 2000. The meaning of
patent citations: report on the NBER/Case Western Research Survey of Patentees. NBER Working Paper 6507.
Jaffe, A.B., Trajtenberg, M., Henderson, R.M., 1993. Geographic
knowledge spillovers as evidenced by patent citations. Quarterly
Journal of Economics 108, 577598.
Klevorick, A.K., Levin, R., Nelson, R., Winter, S., 1994. On the
sources and significance of interindustry differences in technological opportunities. Research Policy 24, 195206.
Link, A.N., Scott, J.T., 2004. U.S. university research parks. Working
Paper. University of North Carolina at Greensboro.
Maddala, G.S., 1983. Limited-Dependent and Qualitative Variables
in Econometrics. Cambridge University Press, Cambridge, UK.
Mansfield, E., 1995. Academic research underlying industrial innovations: sources, characteristics, and financing. Review of Economics and Statistics 77, 5565.
Mian, S.A., 1996. Assessing value-added contributions of university
technology business incubators to tenant firms. Research Policy
25, 325335.
Murray, F., Stern, S., 2004. Do formal intellectual property rights
hinder the free flow of scientific knowledge? Evidence from
patentpaper pairs. Working Paper. MIT.
Phan, P.H., Siegel, D.S., Wright, M., 2004. Science parks and incubators: observations, synthesis and future research. Journal of
Business Venturing 20, 165182.
Rosenwein, R., November 2000. The idea factories. In: Inc.
Rothaermel, F.T., 2001. Complementary assets, strategic alliances,
and the incumbents advantage: an empirical study of industry
and firm effects in the biopharmaceutical industry. Research Policy 30, 12351251.

Rothaermel, F.T., 2002. Technological discontinuities and interfirm


cooperation: what determines a start-ups attractiveness as alliance partner? IEEE Transactions on Engineering Management
49, 388397.
Rothaermel, F.T., Deeds, D.L., 2004. Exploration and exploitation alliances in biotechnology: a system of new product development.
Strategic Management Journal 25, 201221.
Rothaermel, F.T., Hill, C.W.L., 2005. Technological discontinuities
and complementary assets: a longitudinal study of industry and
firm performance. Organization Science 16, 5270.
Rothaermel, F.T, Thursby, M., 2005. Incubator firm failure or graduation? The role of university linkages. Research Policy 34, in
press.
Sampat, B., 2004. Examining patent examination: an analysis of examiner and applicant generated prior art. Working Paper. Georgia
Institute of Technology.
Shane, S., Cable, D., 2002. Network ties, reputation, and the financing of new ventures. Management Science 48, 364381.
Shane, S., Stuart, T., 2002. Organizational endowments and the
performance of university start-ups. Management Science 48,
154170.
Siegel, D.S., Westhead, P., Wright, M., 2003. Assessing the impact of
science parks on the research productivity of firms: exploratory
evidence from the United Kingdom. International Journal of Industrial Organization 21, 1351369.
Stuart, T.E., Hoang, H., Hybels, R.C., 1999. Interorganizational endorsements and the performance of entrepreneurial ventures. Administrative Science Quarterly 44, 315349.
Thursby, J., Thursby, M.C., 2002. Who is selling the ivory tower?
Sources of growth in university licensing. Management Science
48, 90104.
Thursby, J., Thursby, M., 2004. Are faculty critical? Their role in
universityindustry licensing. Contemporary Economic Policy
22, 162178.
Trajtenberg, M., Henderson, R., Jaffe, A., 1997. University versus
corporate patents: a window on the basicness of invention. Economic Innovation and New Technology 5, 1950.
Zucker, L.G., Darby, M.R., 1996. Costly informationfirm transformation, exit, or persistent failure. American Behavioral Scientist
39, 959974.
Zucker, L.G., Darby, M.R., 1998. Intellectual human capital and the
birth of U.S. biotechnology enterprises. American Economic Review 88, 290306.
Zucker, L.G., Darby, M.R., Armstrong, J., 2002. Commercializing
knowledge: university science, knowledge capture, and firm performance in biotechnology. Management Science 48, 138153.